Decoding Trans-Saccadic Prediction Error

Abstract

We are constantly sampling our environment by moving our eyes, but our subjective experience of the world is stable and constant. Stimulus displacement during or shortly after a saccade often goes unnoticed, a phenomenon called the saccadic suppression of displacement. Although we fail to notice such displacements, our oculomotor system computes the prediction errors and adequately adjusts the gaze and future saccadic execution, a phenomenon known as saccadic adaptation. In the present study, we aimed to find a brain signature of the trans-saccadic prediction error that informs the motor system but not explicit perception. We asked participants (either sex) to report whether a visual target was displaced during a saccade while recording electroencephalography (EEG). Using multivariate pattern analysis, we were able to differentiate displacements from no displacements, even when participants failed to report the displacement. In other words, we found that trans-saccadic prediction error is represented in the EEG signal 100 ms after the displacement presentation, mainly in occipital and parieto-occipital channels, even in the absence of explicit perception of the displacement.SIGNIFICANCE STATEMENT Stability in vision occurs even while performing saccades. One suggested mechanism for this counterintuitive visual phenomenon is that external displacement is suppressed during the retinal remapping caused by a saccade. Here, we shed light on the mechanisms of trans-saccadic stability by showing that displacement information is not entirely suppressed and specifically present in the early stages of visual processing. Such a signal is relevant and computed for oculomotor adjustment despite being neglected for perception.

Keywords: EEG; decoding; saccadic suppression of displacement; trans-saccadic error.

Figure 1.

A, Time course of a single trial. The red dot represents the current eye position. After a go signal, participants performed a saccade toward a target placed either on the left or on the right of a central fixation point. At saccade onset the target disappeared and reappeared 50 ms after at the same place (no displacement, 50%) or at a distance d away from the initial position (displacement condition, 25% backward, 25% forward). Participants reported whether the target moved. B, Average distance thresholds d in degrees of visual angle measured in the staircase task. C, Average sensitivity d′ to target displacements. D, Decision criterion. All averages were calculated across participants, and each point represents an individual observer (B–D). E, Individual distributions of landing errors and the box plot of median error across participants. Negative values represent undershoots and positive value overshoots. F, Relationship between post-saccadic landing error and adjustment in subsequent saccadic amplitude. Lines represent their linear relationship for each participant, displacement condition (Backward/No displacement/Forward) and response type (No/Yes). G, Median amplitude residuals of the linear relationship of saccadic adjustment and post-saccadic landing errors separated by displacement condition (Backward, No-displacement, Forward) across participants. Post-saccadic errors were not enough for explaining the saccadic adjustment as the actual adjustment is below/above the expected in the forward/backward conditions, respectively. This intuition can be extracted from F as well, where at no adjustment (y = 0), participants systematically had more positive errors (overshoots) in backward trials than in no-displacement trials, and more negative errors (undershoots) in forward trials. H, Distribution of the median absolute post-saccadic error (PSE) for hits, false alarms (FA), misses, and correct rejections (CR) across participants.

Figure 2.

Decoding displacements from the EEG signal. Left, Dotted vertical lines represent saccade detection (at 0 s) and target reappearance (at 0.05 s). Thick lines represent time points at which classification performance differs from chance (AUC = 0.5). Right, Topographic activation maps at four time windows, with the significant channels highlighted. The color map ranges from −10 (cool colors) to 10 (warm colors) arbitrary units.

Figure 3.

A, Decoding displacements from the eye position data. Dashed lines and shaded areas as in Figure 2. B, Eye position data of participant 1 at 50 ms (target reappearance) and at 600 ms (right before the response go signal). The y-axis is the vertical position on the screen, whereas the x-axis is the distance from the center of the screen, the initial fixation. The plus signs represent the three possible positions of the target across trials (backward displacement, no-displacement, forward displacement). The gaze of participant 1 in all trials is represented by the dots. The dot color represents if there was a displacement (red) or not (blue) in each trial. As in A, for participant 1, there is a clear boundary for classifying the displacement from the eye position at the end of the trial (600 ms) but not at the beginning when the target just reappeared (50 ms), meaning that participant 1 generally performed eye movements toward the target at some point of the trial.